Integrated attack method using non-systematic preprocessing and dynamic weight

The invention provides an integrated attack method using non-systematic preprocessing and dynamic weight. Comprising the following steps: 1, selecting an experimental data set, and determining a model; 2, performing input sample data enhancement; 3, generating an adversarial sample of the integrated model through multi-model dynamic tuning integrated attacks; 4, according to a model weight rebalancing strategy, generating an optimized confrontation sample; step 5, circularly executing the step 3 to the step 4 until the set iteration threshold value is exceeded, and updating the adversarial sample; 6, ending the circulation, and outputting a confrontation sample generated by the final integrated model; and step 7, using the obtained confrontation attack sample to attack the target black box model, and calculating the success rate of the confrontation sample attack model. According to the method, the contribution of each agent model to the attack effect can be monitored and optimized, the migration capability o